1{
  2  "bench_version": 1,
  3  "track": "inverse_gain_control",
  4  "model": "rnn_small",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.003
 10    },
 11    "sweep": [
 12      {
 13        "cfg": {
 14          "lr": 0.001
 15        },
 16        "mean": 5.920182824134827
 17      },
 18      {
 19        "cfg": {
 20          "lr": 0.002
 21        },
 22        "mean": 4.574557304382324
 23      },
 24      {
 25        "cfg": {
 26          "lr": 0.003
 27        },
 28        "mean": 3.198916256427765
 29      }
 30    ],
 31    "full": {
 32      "mean": 3.670998603105545,
 33      "std": 0.63283622936479,
 34      "per_seed": [
 35        3.7461042404174805,
 36        3.104163408279419,
 37        2.9681482315063477,
 38        2.9772491455078125,
 39        4.229879379272461,
 40        4.706507682800293,
 41        4.304690361022949,
 42        3.3312463760375977
 43      ],
 44      "n": 8
 45    }
 46  },
 47  "idea": {
 48    "mean": 1.5069266557693481,
 49    "std": 0.23555139171685704,
 50    "per_seed": [
 51      1.5180333852767944,
 52      1.335729956626892,
 53      1.2053886651992798,
 54      1.2145419120788574,
 55      1.7869166135787964,
 56      1.6817357540130615,
 57      1.8618539571762085,
 58      1.451213002204895
 59    ],
 60    "n": 8
 61  },
 62  "comparison": {
 63    "delta_mean": -2.164071947336197,
 64    "idea_wins": 8,
 65    "n_pairs": 8,
 66    "per_seed_diffs": [
 67      -2.228070855140686,
 68      -1.7684334516525269,
 69      -1.7627595663070679,
 70      -1.762707233428955,
 71      -2.4429627656936646,
 72      -3.0247719287872314,
 73      -2.4428364038467407,
 74      -1.8800333738327026
 75    ],
 76    "p_value": 0.0081,
 77    "mde": 0.3818346490175283,
 78    "mde_rel_pct": 10.40138366422991,
 79    "verdict": "idea better (significant)",
 80    "system_worked": true
 81  },
 82  "custom_track": {
 83    "name": "inverse_gain_control",
 84    "file": "custom_inverse_gain_track.py",
 85    "domain": "dynamics"
 86  },
 87  "idea_sweep": [
 88    {
 89      "cfg": {
 90        "lr": 0.001
 91      },
 92      "mean": 2.0682411789894104
 93    },
 94    {
 95      "cfg": {
 96        "lr": 0.002
 97      },
 98      "mean": 1.5857862532138824
 99    },
100    {
101      "cfg": {
102        "lr": 0.003
103      },
104      "mean": 1.318423479795456
105    }
106  ],
107  "idea_best_cfg": {
108    "lr": 0.003
109  },
110  "protocol_note": "Both systems use identical GRU(3,64), Adam, epochs, batch, data, and the same lr union; only the output parameterization differs.",
111  "mechanism_signature": {
112    "n_test": 400,
113    "observed_action_error_mean_abs": 0.7880314588546753,
114    "predicted_factor_error_mean_abs": 0.7880314588546753,
115    "slope_observed_on_predicted": 1.0,
116    "correlation": 0.9999999999999957,
117    "relative_residual": 9.462353034450643e-08,
118    "inverse_gain_mae": 0.8039304614067078,
119    "prediction": "action_error=(qhat-q_observed)*z",
120    "confirmed": true
121  }
122}